They also said something like that in their late 1995 forecast, at which time, they expected to snatch the $300B gold ring by 2000! Vaulting off the estimated $145B market size

for that cyclical-peak 1995 year, "Semiconductor Industry Faces 5 years of Boring Growth, to Reach $300B in 2000." Well, not exactly. The industry shrank for three years through 1998 ("1998 was a very bad year."), grew back to $205B in 2001, retreated to $139B in 2001 (!!), and has since climbed to about $250B in 2007. Closer to $300B, but not there yet. And, leading to 2007 with six years of growth without a collapse in prices and overall sales, someone might be nervous about where the industry is headed, and how long it can grow before it faces some kind of retreat.

See the history lesson, below:

All that being said, we all make mistakes. As I recall, one Dataquest analyst (c'est moi) said in late 1984, that the 256K DRAM ASP for 1985 would be more than $10, and was roundly booed by the industry for being too aggressive in his pricing decline estimates. It came in about $3.30 for the year; we were all wrong...very wrong. The DRAM industry took a major bath of red ink, and Inmos, Intel, National, Mostek, ITT, Fairchild (!) and some lesser Japanese suppliers, all finally pulled the plug on DRAMs, though only Mostek and perhaps Intel were significant DRAM players at that time; the others were only dabblers.

In a recent website review, Samsung, Micron, and Hynix all show-cased their 2Gb DDR2 product offerings. Big die for sure, and pricey. (In fact, Samsung's first 2Gb was in 80-nm ground rules

and Micron's was in 78nm, even bigger die.) <Samsung's is now its 60-nm process, Hynix in its 66-nm process, and Micron's is in its next generation process. Dense DIMMs for servers need dense chips. We have all seem the difficulty within the industry in bringing the FB DIMM to market, which was specifically designed to solve the "addressable DRAM memory subsystem limitation" that we faced few years ago. Dense DIMMs with next-generations DRAM chips and 2H and 4H stacks, which make for very dense DIMMs, up to 16GB/DIMM, are a way around the signaling issues that arise from having too many smaller DIMMs on a single channel. Sometimes it is worth it to pay exorbitant prices for next generation DRAMs to keep your server product line alive for another year or two.

But these 2Gb DRAMs are just toys for now, as the industry is just now moving in earnest from 512Mb to 1Gb DRAMs, as almost all producers are moving their main production to their 70-75nm nodes. That being the case, it is just a matter time until most DRAMs are +/- 70nm, and 1Gb is the volume runner —look for sometime in 2Q08. With shrinks come higher speed distributions, higher density (and smaller) chips, and lower costs (eventually).

Today the mix of 512Mb DRAMs to 1Gb, in terms of MB shipments, is about 15:1 (July 2007 shipments were 718M of 512M, and 26M of 1Gb). But 1Gb has ramped in line with 70-75nm processing, by 3Q08; the mix can be expected to be about 50-50 for the whole industry. Samsung and Hynix management are now saying they will be 40% 1Gb MB shipment by year-end 2007, and though they are the leaders, it is indicative of how fast the product mix can turnover once key processing milestones and price points are achieved.

For their 2Gb DRAMs, both Samsung and Hynix are offering 800Mb/sec (= 400MHz clock), but so far, Micron, offers only 667. Long traces and overly large die size, plus maybe some design issues are the culprits. With lots of 512M and 1Gb DDR2-1066s around and about, it is certainly only a matter of time before 2Gb's at 800Mb/sec are widely available.

Although there has been little industry noise about Phase Change Memories (PCMs) and no revenue stream, that has not deterred IBM's 'Deep Research'

from another announcement in the NY Times, recasted in the local San Jose Mercury News about IBM's progress in hyper-dense memory technology possibilities of more than 100x times those of current hard drive densities and solid state memories (i.e., NAND flash), to more than 1M MB per square inch. The technology underlying these hyper dense and hyper fast storage media is one its developer Stuart Parkin, 'Father of Giant Magneto Resistive (GMR) technologies for hard drives, calls "Racetrack Memories", which not only can reduce the size of the magnetic storage elements currently seen in HDDs, but can read and write data many hundreds of times faster than existing methods.

His idea, as described in the news article is "to stand billions [!] of ultra fine wire loops around the edge of a silicon chip...and use the electric current to slide infinitesimally small magnets up and down along each of the wires, to be read as digital ones and zeros."

Sounds like science fiction, but maybe not. See:

http://www.nytimes.com/2007/09/11/technology/11storage.html?ref=technology.